Waste Water - UO

Last time, I was confused about which buildings we had obtained waste-water data for. This week, I’ve removed the off-campus site (Robert Guidon Hall) and retained the other 6 on-campus sites (raw data below).


We probably want to summarise our swabs at the weekly (or biweekly?) level for comparison to the waste-water signal. This is what the weekly campus-wide time-series looks like compared with our weekly campus-wide swab positivity:

If we look at the relationship between swabs and waste-water on weeks where both results are available, we see that they have a positive correlation.

The Pearson’s r (linear correlation) is 0.66 (0.38-0.83). The Spearman’s r (rank correlation) is 0.502.. The Kendall’s Tau is 0.36.

Pearson’s r is more affected by outliers. Spearman’s r or Kendall’s Tau are probably the more reliable measures of correlation between these variables.



(Data doesn’t appear normally distributed, btw)



📊 Summary

This plot shows the counts of positive (red) and negative (yellow) samples collected at each facility over time (x-axis). Samples that could not be tested are shown in navy. Only flocked swabs are shown. (Other sponge swabs were collected on 2022-04-28 were for comparing flocked and sponge swabs.)

⡯⡷ Dotplot

This plot shows the counts of positive and negative samples collected at each facility over time.

Model

This section contains results from modeling SARS-CoV-2 cases at UO using swab-PCR results as a predictor.


Specification

We created a random intercepts logistic regression model with the occurrence of cases (binary) as outcome and swab results for the previous week (the proportion of positive swabs) as predictor. The model has a random intercept for each site.

Our model formula is cases ~ positives[lag 1 week] + (1|site).

The model is fit as follows:

swab_model <-
  blme::bglmer(
    cases_binary ~ detection_lag_1week + (1 | site),
    data =  uo_sites,
    family = binomial
  )


Model response time-series

These plots show the swab results, cases, and predictions by the current model.


Model summary

These tables show the model coefficients and statistics.

Model statistics
nobs sigma logLik AIC BIC deviance df.residual
84 1 -24.52 55.04 62.33 42.83 81
Model parameters
effect group term estimate std.error statistic p.value
fixed NA (Intercept) -3.301 0.804 -4.107 0.0000401
fixed NA detection_lag1w 7.114 2.112 3.369 0.0007549
ran_pars site sd__(Intercept) 1.067 NA NA NA


  Relation between UO cases (y/n) and proportion of positive swabs the previous week
Predictors Odds Ratios CI p
(Intercept) 0.04 0.01 – 0.18 <0.001
Swabs @ t-1week 1229.31 19.59 – 77128.28 0.001
Random Effects
σ2 3.29
τ00 site 1.14
ICC 0.26
N site 6
Observations 84
Marginal R2 / Conditional R2 0.378 / 0.538


Model response curves

This plot shows the modelling data as points (detection lagged 1 week and cases at a given site- y/n), as well as how the probability of future cases varies by the previous weeks detection level and site, according to our model (curves).


Random Intercepts for Sites

This plot shows the odds ratios for the intercepts of the model (an intercept for each site).


Swabs and Cases

This panel shows the cases counts at UO over the course of our sampling period. The case data shown represents the days on which a positive test was reported (black rug lines) and the presumed start of transmissiblity for each case (red lines).


Swabs, Wifi, and CO2

This panel shows linked data from swab results, CO2 readings, and wifi traffic (number of users @ peak, daily) during our study period. Unfortunately, we do not have wifi data for 90U.


Wifi Traffic

This panel shows a time-series of the daily peak number of wifi users at UO facilities. Sampling days are highlighted in blue.